Version:  2.0.40 2.2.26 2.4.37 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10

Linux/mm/nommu.c

  1 /*
  2  *  linux/mm/nommu.c
  3  *
  4  *  Replacement code for mm functions to support CPU's that don't
  5  *  have any form of memory management unit (thus no virtual memory).
  6  *
  7  *  See Documentation/nommu-mmap.txt
  8  *
  9  *  Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
 10  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
 11  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
 12  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
 13  *  Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
 14  */
 15 
 16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 17 
 18 #include <linux/export.h>
 19 #include <linux/mm.h>
 20 #include <linux/vmacache.h>
 21 #include <linux/mman.h>
 22 #include <linux/swap.h>
 23 #include <linux/file.h>
 24 #include <linux/highmem.h>
 25 #include <linux/pagemap.h>
 26 #include <linux/slab.h>
 27 #include <linux/vmalloc.h>
 28 #include <linux/blkdev.h>
 29 #include <linux/backing-dev.h>
 30 #include <linux/compiler.h>
 31 #include <linux/mount.h>
 32 #include <linux/personality.h>
 33 #include <linux/security.h>
 34 #include <linux/syscalls.h>
 35 #include <linux/audit.h>
 36 #include <linux/printk.h>
 37 
 38 #include <linux/uaccess.h>
 39 #include <asm/tlb.h>
 40 #include <asm/tlbflush.h>
 41 #include <asm/mmu_context.h>
 42 #include "internal.h"
 43 
 44 void *high_memory;
 45 EXPORT_SYMBOL(high_memory);
 46 struct page *mem_map;
 47 unsigned long max_mapnr;
 48 EXPORT_SYMBOL(max_mapnr);
 49 unsigned long highest_memmap_pfn;
 50 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
 51 int heap_stack_gap = 0;
 52 
 53 atomic_long_t mmap_pages_allocated;
 54 
 55 EXPORT_SYMBOL(mem_map);
 56 
 57 /* list of mapped, potentially shareable regions */
 58 static struct kmem_cache *vm_region_jar;
 59 struct rb_root nommu_region_tree = RB_ROOT;
 60 DECLARE_RWSEM(nommu_region_sem);
 61 
 62 const struct vm_operations_struct generic_file_vm_ops = {
 63 };
 64 
 65 /*
 66  * Return the total memory allocated for this pointer, not
 67  * just what the caller asked for.
 68  *
 69  * Doesn't have to be accurate, i.e. may have races.
 70  */
 71 unsigned int kobjsize(const void *objp)
 72 {
 73         struct page *page;
 74 
 75         /*
 76          * If the object we have should not have ksize performed on it,
 77          * return size of 0
 78          */
 79         if (!objp || !virt_addr_valid(objp))
 80                 return 0;
 81 
 82         page = virt_to_head_page(objp);
 83 
 84         /*
 85          * If the allocator sets PageSlab, we know the pointer came from
 86          * kmalloc().
 87          */
 88         if (PageSlab(page))
 89                 return ksize(objp);
 90 
 91         /*
 92          * If it's not a compound page, see if we have a matching VMA
 93          * region. This test is intentionally done in reverse order,
 94          * so if there's no VMA, we still fall through and hand back
 95          * PAGE_SIZE for 0-order pages.
 96          */
 97         if (!PageCompound(page)) {
 98                 struct vm_area_struct *vma;
 99 
100                 vma = find_vma(current->mm, (unsigned long)objp);
101                 if (vma)
102                         return vma->vm_end - vma->vm_start;
103         }
104 
105         /*
106          * The ksize() function is only guaranteed to work for pointers
107          * returned by kmalloc(). So handle arbitrary pointers here.
108          */
109         return PAGE_SIZE << compound_order(page);
110 }
111 
112 static long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
113                       unsigned long start, unsigned long nr_pages,
114                       unsigned int foll_flags, struct page **pages,
115                       struct vm_area_struct **vmas, int *nonblocking)
116 {
117         struct vm_area_struct *vma;
118         unsigned long vm_flags;
119         int i;
120 
121         /* calculate required read or write permissions.
122          * If FOLL_FORCE is set, we only require the "MAY" flags.
123          */
124         vm_flags  = (foll_flags & FOLL_WRITE) ?
125                         (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
126         vm_flags &= (foll_flags & FOLL_FORCE) ?
127                         (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
128 
129         for (i = 0; i < nr_pages; i++) {
130                 vma = find_vma(mm, start);
131                 if (!vma)
132                         goto finish_or_fault;
133 
134                 /* protect what we can, including chardevs */
135                 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
136                     !(vm_flags & vma->vm_flags))
137                         goto finish_or_fault;
138 
139                 if (pages) {
140                         pages[i] = virt_to_page(start);
141                         if (pages[i])
142                                 get_page(pages[i]);
143                 }
144                 if (vmas)
145                         vmas[i] = vma;
146                 start = (start + PAGE_SIZE) & PAGE_MASK;
147         }
148 
149         return i;
150 
151 finish_or_fault:
152         return i ? : -EFAULT;
153 }
154 
155 /*
156  * get a list of pages in an address range belonging to the specified process
157  * and indicate the VMA that covers each page
158  * - this is potentially dodgy as we may end incrementing the page count of a
159  *   slab page or a secondary page from a compound page
160  * - don't permit access to VMAs that don't support it, such as I/O mappings
161  */
162 long get_user_pages(unsigned long start, unsigned long nr_pages,
163                     unsigned int gup_flags, struct page **pages,
164                     struct vm_area_struct **vmas)
165 {
166         return __get_user_pages(current, current->mm, start, nr_pages,
167                                 gup_flags, pages, vmas, NULL);
168 }
169 EXPORT_SYMBOL(get_user_pages);
170 
171 long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
172                             unsigned int gup_flags, struct page **pages,
173                             int *locked)
174 {
175         return get_user_pages(start, nr_pages, gup_flags, pages, NULL);
176 }
177 EXPORT_SYMBOL(get_user_pages_locked);
178 
179 static long __get_user_pages_unlocked(struct task_struct *tsk,
180                         struct mm_struct *mm, unsigned long start,
181                         unsigned long nr_pages, struct page **pages,
182                         unsigned int gup_flags)
183 {
184         long ret;
185         down_read(&mm->mmap_sem);
186         ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
187                                 NULL, NULL);
188         up_read(&mm->mmap_sem);
189         return ret;
190 }
191 
192 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
193                              struct page **pages, unsigned int gup_flags)
194 {
195         return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
196                                          pages, gup_flags);
197 }
198 EXPORT_SYMBOL(get_user_pages_unlocked);
199 
200 /**
201  * follow_pfn - look up PFN at a user virtual address
202  * @vma: memory mapping
203  * @address: user virtual address
204  * @pfn: location to store found PFN
205  *
206  * Only IO mappings and raw PFN mappings are allowed.
207  *
208  * Returns zero and the pfn at @pfn on success, -ve otherwise.
209  */
210 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
211         unsigned long *pfn)
212 {
213         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
214                 return -EINVAL;
215 
216         *pfn = address >> PAGE_SHIFT;
217         return 0;
218 }
219 EXPORT_SYMBOL(follow_pfn);
220 
221 LIST_HEAD(vmap_area_list);
222 
223 void vfree(const void *addr)
224 {
225         kfree(addr);
226 }
227 EXPORT_SYMBOL(vfree);
228 
229 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
230 {
231         /*
232          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
233          * returns only a logical address.
234          */
235         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
236 }
237 EXPORT_SYMBOL(__vmalloc);
238 
239 void *vmalloc_user(unsigned long size)
240 {
241         void *ret;
242 
243         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
244                         PAGE_KERNEL);
245         if (ret) {
246                 struct vm_area_struct *vma;
247 
248                 down_write(&current->mm->mmap_sem);
249                 vma = find_vma(current->mm, (unsigned long)ret);
250                 if (vma)
251                         vma->vm_flags |= VM_USERMAP;
252                 up_write(&current->mm->mmap_sem);
253         }
254 
255         return ret;
256 }
257 EXPORT_SYMBOL(vmalloc_user);
258 
259 struct page *vmalloc_to_page(const void *addr)
260 {
261         return virt_to_page(addr);
262 }
263 EXPORT_SYMBOL(vmalloc_to_page);
264 
265 unsigned long vmalloc_to_pfn(const void *addr)
266 {
267         return page_to_pfn(virt_to_page(addr));
268 }
269 EXPORT_SYMBOL(vmalloc_to_pfn);
270 
271 long vread(char *buf, char *addr, unsigned long count)
272 {
273         /* Don't allow overflow */
274         if ((unsigned long) buf + count < count)
275                 count = -(unsigned long) buf;
276 
277         memcpy(buf, addr, count);
278         return count;
279 }
280 
281 long vwrite(char *buf, char *addr, unsigned long count)
282 {
283         /* Don't allow overflow */
284         if ((unsigned long) addr + count < count)
285                 count = -(unsigned long) addr;
286 
287         memcpy(addr, buf, count);
288         return count;
289 }
290 
291 /*
292  *      vmalloc  -  allocate virtually contiguous memory
293  *
294  *      @size:          allocation size
295  *
296  *      Allocate enough pages to cover @size from the page level
297  *      allocator and map them into contiguous kernel virtual space.
298  *
299  *      For tight control over page level allocator and protection flags
300  *      use __vmalloc() instead.
301  */
302 void *vmalloc(unsigned long size)
303 {
304        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
305 }
306 EXPORT_SYMBOL(vmalloc);
307 
308 /*
309  *      vzalloc - allocate virtually contiguous memory with zero fill
310  *
311  *      @size:          allocation size
312  *
313  *      Allocate enough pages to cover @size from the page level
314  *      allocator and map them into contiguous kernel virtual space.
315  *      The memory allocated is set to zero.
316  *
317  *      For tight control over page level allocator and protection flags
318  *      use __vmalloc() instead.
319  */
320 void *vzalloc(unsigned long size)
321 {
322         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
323                         PAGE_KERNEL);
324 }
325 EXPORT_SYMBOL(vzalloc);
326 
327 /**
328  * vmalloc_node - allocate memory on a specific node
329  * @size:       allocation size
330  * @node:       numa node
331  *
332  * Allocate enough pages to cover @size from the page level
333  * allocator and map them into contiguous kernel virtual space.
334  *
335  * For tight control over page level allocator and protection flags
336  * use __vmalloc() instead.
337  */
338 void *vmalloc_node(unsigned long size, int node)
339 {
340         return vmalloc(size);
341 }
342 EXPORT_SYMBOL(vmalloc_node);
343 
344 /**
345  * vzalloc_node - allocate memory on a specific node with zero fill
346  * @size:       allocation size
347  * @node:       numa node
348  *
349  * Allocate enough pages to cover @size from the page level
350  * allocator and map them into contiguous kernel virtual space.
351  * The memory allocated is set to zero.
352  *
353  * For tight control over page level allocator and protection flags
354  * use __vmalloc() instead.
355  */
356 void *vzalloc_node(unsigned long size, int node)
357 {
358         return vzalloc(size);
359 }
360 EXPORT_SYMBOL(vzalloc_node);
361 
362 #ifndef PAGE_KERNEL_EXEC
363 # define PAGE_KERNEL_EXEC PAGE_KERNEL
364 #endif
365 
366 /**
367  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
368  *      @size:          allocation size
369  *
370  *      Kernel-internal function to allocate enough pages to cover @size
371  *      the page level allocator and map them into contiguous and
372  *      executable kernel virtual space.
373  *
374  *      For tight control over page level allocator and protection flags
375  *      use __vmalloc() instead.
376  */
377 
378 void *vmalloc_exec(unsigned long size)
379 {
380         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
381 }
382 
383 /**
384  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
385  *      @size:          allocation size
386  *
387  *      Allocate enough 32bit PA addressable pages to cover @size from the
388  *      page level allocator and map them into contiguous kernel virtual space.
389  */
390 void *vmalloc_32(unsigned long size)
391 {
392         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
393 }
394 EXPORT_SYMBOL(vmalloc_32);
395 
396 /**
397  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
398  *      @size:          allocation size
399  *
400  * The resulting memory area is 32bit addressable and zeroed so it can be
401  * mapped to userspace without leaking data.
402  *
403  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
404  * remap_vmalloc_range() are permissible.
405  */
406 void *vmalloc_32_user(unsigned long size)
407 {
408         /*
409          * We'll have to sort out the ZONE_DMA bits for 64-bit,
410          * but for now this can simply use vmalloc_user() directly.
411          */
412         return vmalloc_user(size);
413 }
414 EXPORT_SYMBOL(vmalloc_32_user);
415 
416 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
417 {
418         BUG();
419         return NULL;
420 }
421 EXPORT_SYMBOL(vmap);
422 
423 void vunmap(const void *addr)
424 {
425         BUG();
426 }
427 EXPORT_SYMBOL(vunmap);
428 
429 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
430 {
431         BUG();
432         return NULL;
433 }
434 EXPORT_SYMBOL(vm_map_ram);
435 
436 void vm_unmap_ram(const void *mem, unsigned int count)
437 {
438         BUG();
439 }
440 EXPORT_SYMBOL(vm_unmap_ram);
441 
442 void vm_unmap_aliases(void)
443 {
444 }
445 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
446 
447 /*
448  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
449  * have one.
450  */
451 void __weak vmalloc_sync_all(void)
452 {
453 }
454 
455 /**
456  *      alloc_vm_area - allocate a range of kernel address space
457  *      @size:          size of the area
458  *
459  *      Returns:        NULL on failure, vm_struct on success
460  *
461  *      This function reserves a range of kernel address space, and
462  *      allocates pagetables to map that range.  No actual mappings
463  *      are created.  If the kernel address space is not shared
464  *      between processes, it syncs the pagetable across all
465  *      processes.
466  */
467 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
468 {
469         BUG();
470         return NULL;
471 }
472 EXPORT_SYMBOL_GPL(alloc_vm_area);
473 
474 void free_vm_area(struct vm_struct *area)
475 {
476         BUG();
477 }
478 EXPORT_SYMBOL_GPL(free_vm_area);
479 
480 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
481                    struct page *page)
482 {
483         return -EINVAL;
484 }
485 EXPORT_SYMBOL(vm_insert_page);
486 
487 /*
488  *  sys_brk() for the most part doesn't need the global kernel
489  *  lock, except when an application is doing something nasty
490  *  like trying to un-brk an area that has already been mapped
491  *  to a regular file.  in this case, the unmapping will need
492  *  to invoke file system routines that need the global lock.
493  */
494 SYSCALL_DEFINE1(brk, unsigned long, brk)
495 {
496         struct mm_struct *mm = current->mm;
497 
498         if (brk < mm->start_brk || brk > mm->context.end_brk)
499                 return mm->brk;
500 
501         if (mm->brk == brk)
502                 return mm->brk;
503 
504         /*
505          * Always allow shrinking brk
506          */
507         if (brk <= mm->brk) {
508                 mm->brk = brk;
509                 return brk;
510         }
511 
512         /*
513          * Ok, looks good - let it rip.
514          */
515         flush_icache_range(mm->brk, brk);
516         return mm->brk = brk;
517 }
518 
519 /*
520  * initialise the VMA and region record slabs
521  */
522 void __init mmap_init(void)
523 {
524         int ret;
525 
526         ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
527         VM_BUG_ON(ret);
528         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
529 }
530 
531 /*
532  * validate the region tree
533  * - the caller must hold the region lock
534  */
535 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
536 static noinline void validate_nommu_regions(void)
537 {
538         struct vm_region *region, *last;
539         struct rb_node *p, *lastp;
540 
541         lastp = rb_first(&nommu_region_tree);
542         if (!lastp)
543                 return;
544 
545         last = rb_entry(lastp, struct vm_region, vm_rb);
546         BUG_ON(last->vm_end <= last->vm_start);
547         BUG_ON(last->vm_top < last->vm_end);
548 
549         while ((p = rb_next(lastp))) {
550                 region = rb_entry(p, struct vm_region, vm_rb);
551                 last = rb_entry(lastp, struct vm_region, vm_rb);
552 
553                 BUG_ON(region->vm_end <= region->vm_start);
554                 BUG_ON(region->vm_top < region->vm_end);
555                 BUG_ON(region->vm_start < last->vm_top);
556 
557                 lastp = p;
558         }
559 }
560 #else
561 static void validate_nommu_regions(void)
562 {
563 }
564 #endif
565 
566 /*
567  * add a region into the global tree
568  */
569 static void add_nommu_region(struct vm_region *region)
570 {
571         struct vm_region *pregion;
572         struct rb_node **p, *parent;
573 
574         validate_nommu_regions();
575 
576         parent = NULL;
577         p = &nommu_region_tree.rb_node;
578         while (*p) {
579                 parent = *p;
580                 pregion = rb_entry(parent, struct vm_region, vm_rb);
581                 if (region->vm_start < pregion->vm_start)
582                         p = &(*p)->rb_left;
583                 else if (region->vm_start > pregion->vm_start)
584                         p = &(*p)->rb_right;
585                 else if (pregion == region)
586                         return;
587                 else
588                         BUG();
589         }
590 
591         rb_link_node(&region->vm_rb, parent, p);
592         rb_insert_color(&region->vm_rb, &nommu_region_tree);
593 
594         validate_nommu_regions();
595 }
596 
597 /*
598  * delete a region from the global tree
599  */
600 static void delete_nommu_region(struct vm_region *region)
601 {
602         BUG_ON(!nommu_region_tree.rb_node);
603 
604         validate_nommu_regions();
605         rb_erase(&region->vm_rb, &nommu_region_tree);
606         validate_nommu_regions();
607 }
608 
609 /*
610  * free a contiguous series of pages
611  */
612 static void free_page_series(unsigned long from, unsigned long to)
613 {
614         for (; from < to; from += PAGE_SIZE) {
615                 struct page *page = virt_to_page(from);
616 
617                 atomic_long_dec(&mmap_pages_allocated);
618                 put_page(page);
619         }
620 }
621 
622 /*
623  * release a reference to a region
624  * - the caller must hold the region semaphore for writing, which this releases
625  * - the region may not have been added to the tree yet, in which case vm_top
626  *   will equal vm_start
627  */
628 static void __put_nommu_region(struct vm_region *region)
629         __releases(nommu_region_sem)
630 {
631         BUG_ON(!nommu_region_tree.rb_node);
632 
633         if (--region->vm_usage == 0) {
634                 if (region->vm_top > region->vm_start)
635                         delete_nommu_region(region);
636                 up_write(&nommu_region_sem);
637 
638                 if (region->vm_file)
639                         fput(region->vm_file);
640 
641                 /* IO memory and memory shared directly out of the pagecache
642                  * from ramfs/tmpfs mustn't be released here */
643                 if (region->vm_flags & VM_MAPPED_COPY)
644                         free_page_series(region->vm_start, region->vm_top);
645                 kmem_cache_free(vm_region_jar, region);
646         } else {
647                 up_write(&nommu_region_sem);
648         }
649 }
650 
651 /*
652  * release a reference to a region
653  */
654 static void put_nommu_region(struct vm_region *region)
655 {
656         down_write(&nommu_region_sem);
657         __put_nommu_region(region);
658 }
659 
660 /*
661  * update protection on a vma
662  */
663 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
664 {
665 #ifdef CONFIG_MPU
666         struct mm_struct *mm = vma->vm_mm;
667         long start = vma->vm_start & PAGE_MASK;
668         while (start < vma->vm_end) {
669                 protect_page(mm, start, flags);
670                 start += PAGE_SIZE;
671         }
672         update_protections(mm);
673 #endif
674 }
675 
676 /*
677  * add a VMA into a process's mm_struct in the appropriate place in the list
678  * and tree and add to the address space's page tree also if not an anonymous
679  * page
680  * - should be called with mm->mmap_sem held writelocked
681  */
682 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
683 {
684         struct vm_area_struct *pvma, *prev;
685         struct address_space *mapping;
686         struct rb_node **p, *parent, *rb_prev;
687 
688         BUG_ON(!vma->vm_region);
689 
690         mm->map_count++;
691         vma->vm_mm = mm;
692 
693         protect_vma(vma, vma->vm_flags);
694 
695         /* add the VMA to the mapping */
696         if (vma->vm_file) {
697                 mapping = vma->vm_file->f_mapping;
698 
699                 i_mmap_lock_write(mapping);
700                 flush_dcache_mmap_lock(mapping);
701                 vma_interval_tree_insert(vma, &mapping->i_mmap);
702                 flush_dcache_mmap_unlock(mapping);
703                 i_mmap_unlock_write(mapping);
704         }
705 
706         /* add the VMA to the tree */
707         parent = rb_prev = NULL;
708         p = &mm->mm_rb.rb_node;
709         while (*p) {
710                 parent = *p;
711                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
712 
713                 /* sort by: start addr, end addr, VMA struct addr in that order
714                  * (the latter is necessary as we may get identical VMAs) */
715                 if (vma->vm_start < pvma->vm_start)
716                         p = &(*p)->rb_left;
717                 else if (vma->vm_start > pvma->vm_start) {
718                         rb_prev = parent;
719                         p = &(*p)->rb_right;
720                 } else if (vma->vm_end < pvma->vm_end)
721                         p = &(*p)->rb_left;
722                 else if (vma->vm_end > pvma->vm_end) {
723                         rb_prev = parent;
724                         p = &(*p)->rb_right;
725                 } else if (vma < pvma)
726                         p = &(*p)->rb_left;
727                 else if (vma > pvma) {
728                         rb_prev = parent;
729                         p = &(*p)->rb_right;
730                 } else
731                         BUG();
732         }
733 
734         rb_link_node(&vma->vm_rb, parent, p);
735         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
736 
737         /* add VMA to the VMA list also */
738         prev = NULL;
739         if (rb_prev)
740                 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
741 
742         __vma_link_list(mm, vma, prev, parent);
743 }
744 
745 /*
746  * delete a VMA from its owning mm_struct and address space
747  */
748 static void delete_vma_from_mm(struct vm_area_struct *vma)
749 {
750         int i;
751         struct address_space *mapping;
752         struct mm_struct *mm = vma->vm_mm;
753         struct task_struct *curr = current;
754 
755         protect_vma(vma, 0);
756 
757         mm->map_count--;
758         for (i = 0; i < VMACACHE_SIZE; i++) {
759                 /* if the vma is cached, invalidate the entire cache */
760                 if (curr->vmacache[i] == vma) {
761                         vmacache_invalidate(mm);
762                         break;
763                 }
764         }
765 
766         /* remove the VMA from the mapping */
767         if (vma->vm_file) {
768                 mapping = vma->vm_file->f_mapping;
769 
770                 i_mmap_lock_write(mapping);
771                 flush_dcache_mmap_lock(mapping);
772                 vma_interval_tree_remove(vma, &mapping->i_mmap);
773                 flush_dcache_mmap_unlock(mapping);
774                 i_mmap_unlock_write(mapping);
775         }
776 
777         /* remove from the MM's tree and list */
778         rb_erase(&vma->vm_rb, &mm->mm_rb);
779 
780         if (vma->vm_prev)
781                 vma->vm_prev->vm_next = vma->vm_next;
782         else
783                 mm->mmap = vma->vm_next;
784 
785         if (vma->vm_next)
786                 vma->vm_next->vm_prev = vma->vm_prev;
787 }
788 
789 /*
790  * destroy a VMA record
791  */
792 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
793 {
794         if (vma->vm_ops && vma->vm_ops->close)
795                 vma->vm_ops->close(vma);
796         if (vma->vm_file)
797                 fput(vma->vm_file);
798         put_nommu_region(vma->vm_region);
799         kmem_cache_free(vm_area_cachep, vma);
800 }
801 
802 /*
803  * look up the first VMA in which addr resides, NULL if none
804  * - should be called with mm->mmap_sem at least held readlocked
805  */
806 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
807 {
808         struct vm_area_struct *vma;
809 
810         /* check the cache first */
811         vma = vmacache_find(mm, addr);
812         if (likely(vma))
813                 return vma;
814 
815         /* trawl the list (there may be multiple mappings in which addr
816          * resides) */
817         for (vma = mm->mmap; vma; vma = vma->vm_next) {
818                 if (vma->vm_start > addr)
819                         return NULL;
820                 if (vma->vm_end > addr) {
821                         vmacache_update(addr, vma);
822                         return vma;
823                 }
824         }
825 
826         return NULL;
827 }
828 EXPORT_SYMBOL(find_vma);
829 
830 /*
831  * find a VMA
832  * - we don't extend stack VMAs under NOMMU conditions
833  */
834 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
835 {
836         return find_vma(mm, addr);
837 }
838 
839 /*
840  * expand a stack to a given address
841  * - not supported under NOMMU conditions
842  */
843 int expand_stack(struct vm_area_struct *vma, unsigned long address)
844 {
845         return -ENOMEM;
846 }
847 
848 /*
849  * look up the first VMA exactly that exactly matches addr
850  * - should be called with mm->mmap_sem at least held readlocked
851  */
852 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
853                                              unsigned long addr,
854                                              unsigned long len)
855 {
856         struct vm_area_struct *vma;
857         unsigned long end = addr + len;
858 
859         /* check the cache first */
860         vma = vmacache_find_exact(mm, addr, end);
861         if (vma)
862                 return vma;
863 
864         /* trawl the list (there may be multiple mappings in which addr
865          * resides) */
866         for (vma = mm->mmap; vma; vma = vma->vm_next) {
867                 if (vma->vm_start < addr)
868                         continue;
869                 if (vma->vm_start > addr)
870                         return NULL;
871                 if (vma->vm_end == end) {
872                         vmacache_update(addr, vma);
873                         return vma;
874                 }
875         }
876 
877         return NULL;
878 }
879 
880 /*
881  * determine whether a mapping should be permitted and, if so, what sort of
882  * mapping we're capable of supporting
883  */
884 static int validate_mmap_request(struct file *file,
885                                  unsigned long addr,
886                                  unsigned long len,
887                                  unsigned long prot,
888                                  unsigned long flags,
889                                  unsigned long pgoff,
890                                  unsigned long *_capabilities)
891 {
892         unsigned long capabilities, rlen;
893         int ret;
894 
895         /* do the simple checks first */
896         if (flags & MAP_FIXED)
897                 return -EINVAL;
898 
899         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
900             (flags & MAP_TYPE) != MAP_SHARED)
901                 return -EINVAL;
902 
903         if (!len)
904                 return -EINVAL;
905 
906         /* Careful about overflows.. */
907         rlen = PAGE_ALIGN(len);
908         if (!rlen || rlen > TASK_SIZE)
909                 return -ENOMEM;
910 
911         /* offset overflow? */
912         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
913                 return -EOVERFLOW;
914 
915         if (file) {
916                 /* files must support mmap */
917                 if (!file->f_op->mmap)
918                         return -ENODEV;
919 
920                 /* work out if what we've got could possibly be shared
921                  * - we support chardevs that provide their own "memory"
922                  * - we support files/blockdevs that are memory backed
923                  */
924                 if (file->f_op->mmap_capabilities) {
925                         capabilities = file->f_op->mmap_capabilities(file);
926                 } else {
927                         /* no explicit capabilities set, so assume some
928                          * defaults */
929                         switch (file_inode(file)->i_mode & S_IFMT) {
930                         case S_IFREG:
931                         case S_IFBLK:
932                                 capabilities = NOMMU_MAP_COPY;
933                                 break;
934 
935                         case S_IFCHR:
936                                 capabilities =
937                                         NOMMU_MAP_DIRECT |
938                                         NOMMU_MAP_READ |
939                                         NOMMU_MAP_WRITE;
940                                 break;
941 
942                         default:
943                                 return -EINVAL;
944                         }
945                 }
946 
947                 /* eliminate any capabilities that we can't support on this
948                  * device */
949                 if (!file->f_op->get_unmapped_area)
950                         capabilities &= ~NOMMU_MAP_DIRECT;
951                 if (!(file->f_mode & FMODE_CAN_READ))
952                         capabilities &= ~NOMMU_MAP_COPY;
953 
954                 /* The file shall have been opened with read permission. */
955                 if (!(file->f_mode & FMODE_READ))
956                         return -EACCES;
957 
958                 if (flags & MAP_SHARED) {
959                         /* do checks for writing, appending and locking */
960                         if ((prot & PROT_WRITE) &&
961                             !(file->f_mode & FMODE_WRITE))
962                                 return -EACCES;
963 
964                         if (IS_APPEND(file_inode(file)) &&
965                             (file->f_mode & FMODE_WRITE))
966                                 return -EACCES;
967 
968                         if (locks_verify_locked(file))
969                                 return -EAGAIN;
970 
971                         if (!(capabilities & NOMMU_MAP_DIRECT))
972                                 return -ENODEV;
973 
974                         /* we mustn't privatise shared mappings */
975                         capabilities &= ~NOMMU_MAP_COPY;
976                 } else {
977                         /* we're going to read the file into private memory we
978                          * allocate */
979                         if (!(capabilities & NOMMU_MAP_COPY))
980                                 return -ENODEV;
981 
982                         /* we don't permit a private writable mapping to be
983                          * shared with the backing device */
984                         if (prot & PROT_WRITE)
985                                 capabilities &= ~NOMMU_MAP_DIRECT;
986                 }
987 
988                 if (capabilities & NOMMU_MAP_DIRECT) {
989                         if (((prot & PROT_READ)  && !(capabilities & NOMMU_MAP_READ))  ||
990                             ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
991                             ((prot & PROT_EXEC)  && !(capabilities & NOMMU_MAP_EXEC))
992                             ) {
993                                 capabilities &= ~NOMMU_MAP_DIRECT;
994                                 if (flags & MAP_SHARED) {
995                                         pr_warn("MAP_SHARED not completely supported on !MMU\n");
996                                         return -EINVAL;
997                                 }
998                         }
999                 }
1000 
1001                 /* handle executable mappings and implied executable
1002                  * mappings */
1003                 if (path_noexec(&file->f_path)) {
1004                         if (prot & PROT_EXEC)
1005                                 return -EPERM;
1006                 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1007                         /* handle implication of PROT_EXEC by PROT_READ */
1008                         if (current->personality & READ_IMPLIES_EXEC) {
1009                                 if (capabilities & NOMMU_MAP_EXEC)
1010                                         prot |= PROT_EXEC;
1011                         }
1012                 } else if ((prot & PROT_READ) &&
1013                          (prot & PROT_EXEC) &&
1014                          !(capabilities & NOMMU_MAP_EXEC)
1015                          ) {
1016                         /* backing file is not executable, try to copy */
1017                         capabilities &= ~NOMMU_MAP_DIRECT;
1018                 }
1019         } else {
1020                 /* anonymous mappings are always memory backed and can be
1021                  * privately mapped
1022                  */
1023                 capabilities = NOMMU_MAP_COPY;
1024 
1025                 /* handle PROT_EXEC implication by PROT_READ */
1026                 if ((prot & PROT_READ) &&
1027                     (current->personality & READ_IMPLIES_EXEC))
1028                         prot |= PROT_EXEC;
1029         }
1030 
1031         /* allow the security API to have its say */
1032         ret = security_mmap_addr(addr);
1033         if (ret < 0)
1034                 return ret;
1035 
1036         /* looks okay */
1037         *_capabilities = capabilities;
1038         return 0;
1039 }
1040 
1041 /*
1042  * we've determined that we can make the mapping, now translate what we
1043  * now know into VMA flags
1044  */
1045 static unsigned long determine_vm_flags(struct file *file,
1046                                         unsigned long prot,
1047                                         unsigned long flags,
1048                                         unsigned long capabilities)
1049 {
1050         unsigned long vm_flags;
1051 
1052         vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
1053         /* vm_flags |= mm->def_flags; */
1054 
1055         if (!(capabilities & NOMMU_MAP_DIRECT)) {
1056                 /* attempt to share read-only copies of mapped file chunks */
1057                 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1058                 if (file && !(prot & PROT_WRITE))
1059                         vm_flags |= VM_MAYSHARE;
1060         } else {
1061                 /* overlay a shareable mapping on the backing device or inode
1062                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1063                  * romfs/cramfs */
1064                 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1065                 if (flags & MAP_SHARED)
1066                         vm_flags |= VM_SHARED;
1067         }
1068 
1069         /* refuse to let anyone share private mappings with this process if
1070          * it's being traced - otherwise breakpoints set in it may interfere
1071          * with another untraced process
1072          */
1073         if ((flags & MAP_PRIVATE) && current->ptrace)
1074                 vm_flags &= ~VM_MAYSHARE;
1075 
1076         return vm_flags;
1077 }
1078 
1079 /*
1080  * set up a shared mapping on a file (the driver or filesystem provides and
1081  * pins the storage)
1082  */
1083 static int do_mmap_shared_file(struct vm_area_struct *vma)
1084 {
1085         int ret;
1086 
1087         ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1088         if (ret == 0) {
1089                 vma->vm_region->vm_top = vma->vm_region->vm_end;
1090                 return 0;
1091         }
1092         if (ret != -ENOSYS)
1093                 return ret;
1094 
1095         /* getting -ENOSYS indicates that direct mmap isn't possible (as
1096          * opposed to tried but failed) so we can only give a suitable error as
1097          * it's not possible to make a private copy if MAP_SHARED was given */
1098         return -ENODEV;
1099 }
1100 
1101 /*
1102  * set up a private mapping or an anonymous shared mapping
1103  */
1104 static int do_mmap_private(struct vm_area_struct *vma,
1105                            struct vm_region *region,
1106                            unsigned long len,
1107                            unsigned long capabilities)
1108 {
1109         unsigned long total, point;
1110         void *base;
1111         int ret, order;
1112 
1113         /* invoke the file's mapping function so that it can keep track of
1114          * shared mappings on devices or memory
1115          * - VM_MAYSHARE will be set if it may attempt to share
1116          */
1117         if (capabilities & NOMMU_MAP_DIRECT) {
1118                 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1119                 if (ret == 0) {
1120                         /* shouldn't return success if we're not sharing */
1121                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1122                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1123                         return 0;
1124                 }
1125                 if (ret != -ENOSYS)
1126                         return ret;
1127 
1128                 /* getting an ENOSYS error indicates that direct mmap isn't
1129                  * possible (as opposed to tried but failed) so we'll try to
1130                  * make a private copy of the data and map that instead */
1131         }
1132 
1133 
1134         /* allocate some memory to hold the mapping
1135          * - note that this may not return a page-aligned address if the object
1136          *   we're allocating is smaller than a page
1137          */
1138         order = get_order(len);
1139         total = 1 << order;
1140         point = len >> PAGE_SHIFT;
1141 
1142         /* we don't want to allocate a power-of-2 sized page set */
1143         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1144                 total = point;
1145 
1146         base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1147         if (!base)
1148                 goto enomem;
1149 
1150         atomic_long_add(total, &mmap_pages_allocated);
1151 
1152         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1153         region->vm_start = (unsigned long) base;
1154         region->vm_end   = region->vm_start + len;
1155         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1156 
1157         vma->vm_start = region->vm_start;
1158         vma->vm_end   = region->vm_start + len;
1159 
1160         if (vma->vm_file) {
1161                 /* read the contents of a file into the copy */
1162                 mm_segment_t old_fs;
1163                 loff_t fpos;
1164 
1165                 fpos = vma->vm_pgoff;
1166                 fpos <<= PAGE_SHIFT;
1167 
1168                 old_fs = get_fs();
1169                 set_fs(KERNEL_DS);
1170                 ret = __vfs_read(vma->vm_file, base, len, &fpos);
1171                 set_fs(old_fs);
1172 
1173                 if (ret < 0)
1174                         goto error_free;
1175 
1176                 /* clear the last little bit */
1177                 if (ret < len)
1178                         memset(base + ret, 0, len - ret);
1179 
1180         }
1181 
1182         return 0;
1183 
1184 error_free:
1185         free_page_series(region->vm_start, region->vm_top);
1186         region->vm_start = vma->vm_start = 0;
1187         region->vm_end   = vma->vm_end = 0;
1188         region->vm_top   = 0;
1189         return ret;
1190 
1191 enomem:
1192         pr_err("Allocation of length %lu from process %d (%s) failed\n",
1193                len, current->pid, current->comm);
1194         show_free_areas(0);
1195         return -ENOMEM;
1196 }
1197 
1198 /*
1199  * handle mapping creation for uClinux
1200  */
1201 unsigned long do_mmap(struct file *file,
1202                         unsigned long addr,
1203                         unsigned long len,
1204                         unsigned long prot,
1205                         unsigned long flags,
1206                         vm_flags_t vm_flags,
1207                         unsigned long pgoff,
1208                         unsigned long *populate)
1209 {
1210         struct vm_area_struct *vma;
1211         struct vm_region *region;
1212         struct rb_node *rb;
1213         unsigned long capabilities, result;
1214         int ret;
1215 
1216         *populate = 0;
1217 
1218         /* decide whether we should attempt the mapping, and if so what sort of
1219          * mapping */
1220         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1221                                     &capabilities);
1222         if (ret < 0)
1223                 return ret;
1224 
1225         /* we ignore the address hint */
1226         addr = 0;
1227         len = PAGE_ALIGN(len);
1228 
1229         /* we've determined that we can make the mapping, now translate what we
1230          * now know into VMA flags */
1231         vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1232 
1233         /* we're going to need to record the mapping */
1234         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1235         if (!region)
1236                 goto error_getting_region;
1237 
1238         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1239         if (!vma)
1240                 goto error_getting_vma;
1241 
1242         region->vm_usage = 1;
1243         region->vm_flags = vm_flags;
1244         region->vm_pgoff = pgoff;
1245 
1246         INIT_LIST_HEAD(&vma->anon_vma_chain);
1247         vma->vm_flags = vm_flags;
1248         vma->vm_pgoff = pgoff;
1249 
1250         if (file) {
1251                 region->vm_file = get_file(file);
1252                 vma->vm_file = get_file(file);
1253         }
1254 
1255         down_write(&nommu_region_sem);
1256 
1257         /* if we want to share, we need to check for regions created by other
1258          * mmap() calls that overlap with our proposed mapping
1259          * - we can only share with a superset match on most regular files
1260          * - shared mappings on character devices and memory backed files are
1261          *   permitted to overlap inexactly as far as we are concerned for in
1262          *   these cases, sharing is handled in the driver or filesystem rather
1263          *   than here
1264          */
1265         if (vm_flags & VM_MAYSHARE) {
1266                 struct vm_region *pregion;
1267                 unsigned long pglen, rpglen, pgend, rpgend, start;
1268 
1269                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1270                 pgend = pgoff + pglen;
1271 
1272                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1273                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1274 
1275                         if (!(pregion->vm_flags & VM_MAYSHARE))
1276                                 continue;
1277 
1278                         /* search for overlapping mappings on the same file */
1279                         if (file_inode(pregion->vm_file) !=
1280                             file_inode(file))
1281                                 continue;
1282 
1283                         if (pregion->vm_pgoff >= pgend)
1284                                 continue;
1285 
1286                         rpglen = pregion->vm_end - pregion->vm_start;
1287                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1288                         rpgend = pregion->vm_pgoff + rpglen;
1289                         if (pgoff >= rpgend)
1290                                 continue;
1291 
1292                         /* handle inexactly overlapping matches between
1293                          * mappings */
1294                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1295                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1296                                 /* new mapping is not a subset of the region */
1297                                 if (!(capabilities & NOMMU_MAP_DIRECT))
1298                                         goto sharing_violation;
1299                                 continue;
1300                         }
1301 
1302                         /* we've found a region we can share */
1303                         pregion->vm_usage++;
1304                         vma->vm_region = pregion;
1305                         start = pregion->vm_start;
1306                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1307                         vma->vm_start = start;
1308                         vma->vm_end = start + len;
1309 
1310                         if (pregion->vm_flags & VM_MAPPED_COPY)
1311                                 vma->vm_flags |= VM_MAPPED_COPY;
1312                         else {
1313                                 ret = do_mmap_shared_file(vma);
1314                                 if (ret < 0) {
1315                                         vma->vm_region = NULL;
1316                                         vma->vm_start = 0;
1317                                         vma->vm_end = 0;
1318                                         pregion->vm_usage--;
1319                                         pregion = NULL;
1320                                         goto error_just_free;
1321                                 }
1322                         }
1323                         fput(region->vm_file);
1324                         kmem_cache_free(vm_region_jar, region);
1325                         region = pregion;
1326                         result = start;
1327                         goto share;
1328                 }
1329 
1330                 /* obtain the address at which to make a shared mapping
1331                  * - this is the hook for quasi-memory character devices to
1332                  *   tell us the location of a shared mapping
1333                  */
1334                 if (capabilities & NOMMU_MAP_DIRECT) {
1335                         addr = file->f_op->get_unmapped_area(file, addr, len,
1336                                                              pgoff, flags);
1337                         if (IS_ERR_VALUE(addr)) {
1338                                 ret = addr;
1339                                 if (ret != -ENOSYS)
1340                                         goto error_just_free;
1341 
1342                                 /* the driver refused to tell us where to site
1343                                  * the mapping so we'll have to attempt to copy
1344                                  * it */
1345                                 ret = -ENODEV;
1346                                 if (!(capabilities & NOMMU_MAP_COPY))
1347                                         goto error_just_free;
1348 
1349                                 capabilities &= ~NOMMU_MAP_DIRECT;
1350                         } else {
1351                                 vma->vm_start = region->vm_start = addr;
1352                                 vma->vm_end = region->vm_end = addr + len;
1353                         }
1354                 }
1355         }
1356 
1357         vma->vm_region = region;
1358 
1359         /* set up the mapping
1360          * - the region is filled in if NOMMU_MAP_DIRECT is still set
1361          */
1362         if (file && vma->vm_flags & VM_SHARED)
1363                 ret = do_mmap_shared_file(vma);
1364         else
1365                 ret = do_mmap_private(vma, region, len, capabilities);
1366         if (ret < 0)
1367                 goto error_just_free;
1368         add_nommu_region(region);
1369 
1370         /* clear anonymous mappings that don't ask for uninitialized data */
1371         if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1372                 memset((void *)region->vm_start, 0,
1373                        region->vm_end - region->vm_start);
1374 
1375         /* okay... we have a mapping; now we have to register it */
1376         result = vma->vm_start;
1377 
1378         current->mm->total_vm += len >> PAGE_SHIFT;
1379 
1380 share:
1381         add_vma_to_mm(current->mm, vma);
1382 
1383         /* we flush the region from the icache only when the first executable
1384          * mapping of it is made  */
1385         if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1386                 flush_icache_range(region->vm_start, region->vm_end);
1387                 region->vm_icache_flushed = true;
1388         }
1389 
1390         up_write(&nommu_region_sem);
1391 
1392         return result;
1393 
1394 error_just_free:
1395         up_write(&nommu_region_sem);
1396 error:
1397         if (region->vm_file)
1398                 fput(region->vm_file);
1399         kmem_cache_free(vm_region_jar, region);
1400         if (vma->vm_file)
1401                 fput(vma->vm_file);
1402         kmem_cache_free(vm_area_cachep, vma);
1403         return ret;
1404 
1405 sharing_violation:
1406         up_write(&nommu_region_sem);
1407         pr_warn("Attempt to share mismatched mappings\n");
1408         ret = -EINVAL;
1409         goto error;
1410 
1411 error_getting_vma:
1412         kmem_cache_free(vm_region_jar, region);
1413         pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1414                         len, current->pid);
1415         show_free_areas(0);
1416         return -ENOMEM;
1417 
1418 error_getting_region:
1419         pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1420                         len, current->pid);
1421         show_free_areas(0);
1422         return -ENOMEM;
1423 }
1424 
1425 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1426                 unsigned long, prot, unsigned long, flags,
1427                 unsigned long, fd, unsigned long, pgoff)
1428 {
1429         struct file *file = NULL;
1430         unsigned long retval = -EBADF;
1431 
1432         audit_mmap_fd(fd, flags);
1433         if (!(flags & MAP_ANONYMOUS)) {
1434                 file = fget(fd);
1435                 if (!file)
1436                         goto out;
1437         }
1438 
1439         flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1440 
1441         retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1442 
1443         if (file)
1444                 fput(file);
1445 out:
1446         return retval;
1447 }
1448 
1449 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1450 struct mmap_arg_struct {
1451         unsigned long addr;
1452         unsigned long len;
1453         unsigned long prot;
1454         unsigned long flags;
1455         unsigned long fd;
1456         unsigned long offset;
1457 };
1458 
1459 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1460 {
1461         struct mmap_arg_struct a;
1462 
1463         if (copy_from_user(&a, arg, sizeof(a)))
1464                 return -EFAULT;
1465         if (offset_in_page(a.offset))
1466                 return -EINVAL;
1467 
1468         return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1469                               a.offset >> PAGE_SHIFT);
1470 }
1471 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1472 
1473 /*
1474  * split a vma into two pieces at address 'addr', a new vma is allocated either
1475  * for the first part or the tail.
1476  */
1477 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1478               unsigned long addr, int new_below)
1479 {
1480         struct vm_area_struct *new;
1481         struct vm_region *region;
1482         unsigned long npages;
1483 
1484         /* we're only permitted to split anonymous regions (these should have
1485          * only a single usage on the region) */
1486         if (vma->vm_file)
1487                 return -ENOMEM;
1488 
1489         if (mm->map_count >= sysctl_max_map_count)
1490                 return -ENOMEM;
1491 
1492         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1493         if (!region)
1494                 return -ENOMEM;
1495 
1496         new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1497         if (!new) {
1498                 kmem_cache_free(vm_region_jar, region);
1499                 return -ENOMEM;
1500         }
1501 
1502         /* most fields are the same, copy all, and then fixup */
1503         *new = *vma;
1504         *region = *vma->vm_region;
1505         new->vm_region = region;
1506 
1507         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1508 
1509         if (new_below) {
1510                 region->vm_top = region->vm_end = new->vm_end = addr;
1511         } else {
1512                 region->vm_start = new->vm_start = addr;
1513                 region->vm_pgoff = new->vm_pgoff += npages;
1514         }
1515 
1516         if (new->vm_ops && new->vm_ops->open)
1517                 new->vm_ops->open(new);
1518 
1519         delete_vma_from_mm(vma);
1520         down_write(&nommu_region_sem);
1521         delete_nommu_region(vma->vm_region);
1522         if (new_below) {
1523                 vma->vm_region->vm_start = vma->vm_start = addr;
1524                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1525         } else {
1526                 vma->vm_region->vm_end = vma->vm_end = addr;
1527                 vma->vm_region->vm_top = addr;
1528         }
1529         add_nommu_region(vma->vm_region);
1530         add_nommu_region(new->vm_region);
1531         up_write(&nommu_region_sem);
1532         add_vma_to_mm(mm, vma);
1533         add_vma_to_mm(mm, new);
1534         return 0;
1535 }
1536 
1537 /*
1538  * shrink a VMA by removing the specified chunk from either the beginning or
1539  * the end
1540  */
1541 static int shrink_vma(struct mm_struct *mm,
1542                       struct vm_area_struct *vma,
1543                       unsigned long from, unsigned long to)
1544 {
1545         struct vm_region *region;
1546 
1547         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1548          * and list */
1549         delete_vma_from_mm(vma);
1550         if (from > vma->vm_start)
1551                 vma->vm_end = from;
1552         else
1553                 vma->vm_start = to;
1554         add_vma_to_mm(mm, vma);
1555 
1556         /* cut the backing region down to size */
1557         region = vma->vm_region;
1558         BUG_ON(region->vm_usage != 1);
1559 
1560         down_write(&nommu_region_sem);
1561         delete_nommu_region(region);
1562         if (from > region->vm_start) {
1563                 to = region->vm_top;
1564                 region->vm_top = region->vm_end = from;
1565         } else {
1566                 region->vm_start = to;
1567         }
1568         add_nommu_region(region);
1569         up_write(&nommu_region_sem);
1570 
1571         free_page_series(from, to);
1572         return 0;
1573 }
1574 
1575 /*
1576  * release a mapping
1577  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1578  *   VMA, though it need not cover the whole VMA
1579  */
1580 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1581 {
1582         struct vm_area_struct *vma;
1583         unsigned long end;
1584         int ret;
1585 
1586         len = PAGE_ALIGN(len);
1587         if (len == 0)
1588                 return -EINVAL;
1589 
1590         end = start + len;
1591 
1592         /* find the first potentially overlapping VMA */
1593         vma = find_vma(mm, start);
1594         if (!vma) {
1595                 static int limit;
1596                 if (limit < 5) {
1597                         pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1598                                         current->pid, current->comm,
1599                                         start, start + len - 1);
1600                         limit++;
1601                 }
1602                 return -EINVAL;
1603         }
1604 
1605         /* we're allowed to split an anonymous VMA but not a file-backed one */
1606         if (vma->vm_file) {
1607                 do {
1608                         if (start > vma->vm_start)
1609                                 return -EINVAL;
1610                         if (end == vma->vm_end)
1611                                 goto erase_whole_vma;
1612                         vma = vma->vm_next;
1613                 } while (vma);
1614                 return -EINVAL;
1615         } else {
1616                 /* the chunk must be a subset of the VMA found */
1617                 if (start == vma->vm_start && end == vma->vm_end)
1618                         goto erase_whole_vma;
1619                 if (start < vma->vm_start || end > vma->vm_end)
1620                         return -EINVAL;
1621                 if (offset_in_page(start))
1622                         return -EINVAL;
1623                 if (end != vma->vm_end && offset_in_page(end))
1624                         return -EINVAL;
1625                 if (start != vma->vm_start && end != vma->vm_end) {
1626                         ret = split_vma(mm, vma, start, 1);
1627                         if (ret < 0)
1628                                 return ret;
1629                 }
1630                 return shrink_vma(mm, vma, start, end);
1631         }
1632 
1633 erase_whole_vma:
1634         delete_vma_from_mm(vma);
1635         delete_vma(mm, vma);
1636         return 0;
1637 }
1638 EXPORT_SYMBOL(do_munmap);
1639 
1640 int vm_munmap(unsigned long addr, size_t len)
1641 {
1642         struct mm_struct *mm = current->mm;
1643         int ret;
1644 
1645         down_write(&mm->mmap_sem);
1646         ret = do_munmap(mm, addr, len);
1647         up_write(&mm->mmap_sem);
1648         return ret;
1649 }
1650 EXPORT_SYMBOL(vm_munmap);
1651 
1652 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1653 {
1654         return vm_munmap(addr, len);
1655 }
1656 
1657 /*
1658  * release all the mappings made in a process's VM space
1659  */
1660 void exit_mmap(struct mm_struct *mm)
1661 {
1662         struct vm_area_struct *vma;
1663 
1664         if (!mm)
1665                 return;
1666 
1667         mm->total_vm = 0;
1668 
1669         while ((vma = mm->mmap)) {
1670                 mm->mmap = vma->vm_next;
1671                 delete_vma_from_mm(vma);
1672                 delete_vma(mm, vma);
1673                 cond_resched();
1674         }
1675 }
1676 
1677 int vm_brk(unsigned long addr, unsigned long len)
1678 {
1679         return -ENOMEM;
1680 }
1681 
1682 /*
1683  * expand (or shrink) an existing mapping, potentially moving it at the same
1684  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1685  *
1686  * under NOMMU conditions, we only permit changing a mapping's size, and only
1687  * as long as it stays within the region allocated by do_mmap_private() and the
1688  * block is not shareable
1689  *
1690  * MREMAP_FIXED is not supported under NOMMU conditions
1691  */
1692 static unsigned long do_mremap(unsigned long addr,
1693                         unsigned long old_len, unsigned long new_len,
1694                         unsigned long flags, unsigned long new_addr)
1695 {
1696         struct vm_area_struct *vma;
1697 
1698         /* insanity checks first */
1699         old_len = PAGE_ALIGN(old_len);
1700         new_len = PAGE_ALIGN(new_len);
1701         if (old_len == 0 || new_len == 0)
1702                 return (unsigned long) -EINVAL;
1703 
1704         if (offset_in_page(addr))
1705                 return -EINVAL;
1706 
1707         if (flags & MREMAP_FIXED && new_addr != addr)
1708                 return (unsigned long) -EINVAL;
1709 
1710         vma = find_vma_exact(current->mm, addr, old_len);
1711         if (!vma)
1712                 return (unsigned long) -EINVAL;
1713 
1714         if (vma->vm_end != vma->vm_start + old_len)
1715                 return (unsigned long) -EFAULT;
1716 
1717         if (vma->vm_flags & VM_MAYSHARE)
1718                 return (unsigned long) -EPERM;
1719 
1720         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1721                 return (unsigned long) -ENOMEM;
1722 
1723         /* all checks complete - do it */
1724         vma->vm_end = vma->vm_start + new_len;
1725         return vma->vm_start;
1726 }
1727 
1728 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1729                 unsigned long, new_len, unsigned long, flags,
1730                 unsigned long, new_addr)
1731 {
1732         unsigned long ret;
1733 
1734         down_write(&current->mm->mmap_sem);
1735         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1736         up_write(&current->mm->mmap_sem);
1737         return ret;
1738 }
1739 
1740 struct page *follow_page_mask(struct vm_area_struct *vma,
1741                               unsigned long address, unsigned int flags,
1742                               unsigned int *page_mask)
1743 {
1744         *page_mask = 0;
1745         return NULL;
1746 }
1747 
1748 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1749                 unsigned long pfn, unsigned long size, pgprot_t prot)
1750 {
1751         if (addr != (pfn << PAGE_SHIFT))
1752                 return -EINVAL;
1753 
1754         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1755         return 0;
1756 }
1757 EXPORT_SYMBOL(remap_pfn_range);
1758 
1759 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1760 {
1761         unsigned long pfn = start >> PAGE_SHIFT;
1762         unsigned long vm_len = vma->vm_end - vma->vm_start;
1763 
1764         pfn += vma->vm_pgoff;
1765         return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1766 }
1767 EXPORT_SYMBOL(vm_iomap_memory);
1768 
1769 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1770                         unsigned long pgoff)
1771 {
1772         unsigned int size = vma->vm_end - vma->vm_start;
1773 
1774         if (!(vma->vm_flags & VM_USERMAP))
1775                 return -EINVAL;
1776 
1777         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1778         vma->vm_end = vma->vm_start + size;
1779 
1780         return 0;
1781 }
1782 EXPORT_SYMBOL(remap_vmalloc_range);
1783 
1784 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1785         unsigned long len, unsigned long pgoff, unsigned long flags)
1786 {
1787         return -ENOMEM;
1788 }
1789 
1790 void unmap_mapping_range(struct address_space *mapping,
1791                          loff_t const holebegin, loff_t const holelen,
1792                          int even_cows)
1793 {
1794 }
1795 EXPORT_SYMBOL(unmap_mapping_range);
1796 
1797 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1798 {
1799         BUG();
1800         return 0;
1801 }
1802 EXPORT_SYMBOL(filemap_fault);
1803 
1804 void filemap_map_pages(struct vm_fault *vmf,
1805                 pgoff_t start_pgoff, pgoff_t end_pgoff)
1806 {
1807         BUG();
1808 }
1809 EXPORT_SYMBOL(filemap_map_pages);
1810 
1811 int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1812                 unsigned long addr, void *buf, int len, unsigned int gup_flags)
1813 {
1814         struct vm_area_struct *vma;
1815         int write = gup_flags & FOLL_WRITE;
1816 
1817         down_read(&mm->mmap_sem);
1818 
1819         /* the access must start within one of the target process's mappings */
1820         vma = find_vma(mm, addr);
1821         if (vma) {
1822                 /* don't overrun this mapping */
1823                 if (addr + len >= vma->vm_end)
1824                         len = vma->vm_end - addr;
1825 
1826                 /* only read or write mappings where it is permitted */
1827                 if (write && vma->vm_flags & VM_MAYWRITE)
1828                         copy_to_user_page(vma, NULL, addr,
1829                                          (void *) addr, buf, len);
1830                 else if (!write && vma->vm_flags & VM_MAYREAD)
1831                         copy_from_user_page(vma, NULL, addr,
1832                                             buf, (void *) addr, len);
1833                 else
1834                         len = 0;
1835         } else {
1836                 len = 0;
1837         }
1838 
1839         up_read(&mm->mmap_sem);
1840 
1841         return len;
1842 }
1843 
1844 /**
1845  * @access_remote_vm - access another process' address space
1846  * @mm:         the mm_struct of the target address space
1847  * @addr:       start address to access
1848  * @buf:        source or destination buffer
1849  * @len:        number of bytes to transfer
1850  * @gup_flags:  flags modifying lookup behaviour
1851  *
1852  * The caller must hold a reference on @mm.
1853  */
1854 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1855                 void *buf, int len, unsigned int gup_flags)
1856 {
1857         return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
1858 }
1859 
1860 /*
1861  * Access another process' address space.
1862  * - source/target buffer must be kernel space
1863  */
1864 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1865                 unsigned int gup_flags)
1866 {
1867         struct mm_struct *mm;
1868 
1869         if (addr + len < addr)
1870                 return 0;
1871 
1872         mm = get_task_mm(tsk);
1873         if (!mm)
1874                 return 0;
1875 
1876         len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
1877 
1878         mmput(mm);
1879         return len;
1880 }
1881 EXPORT_SYMBOL_GPL(access_process_vm);
1882 
1883 /**
1884  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1885  * @inode: The inode to check
1886  * @size: The current filesize of the inode
1887  * @newsize: The proposed filesize of the inode
1888  *
1889  * Check the shared mappings on an inode on behalf of a shrinking truncate to
1890  * make sure that that any outstanding VMAs aren't broken and then shrink the
1891  * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1892  * automatically grant mappings that are too large.
1893  */
1894 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1895                                 size_t newsize)
1896 {
1897         struct vm_area_struct *vma;
1898         struct vm_region *region;
1899         pgoff_t low, high;
1900         size_t r_size, r_top;
1901 
1902         low = newsize >> PAGE_SHIFT;
1903         high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1904 
1905         down_write(&nommu_region_sem);
1906         i_mmap_lock_read(inode->i_mapping);
1907 
1908         /* search for VMAs that fall within the dead zone */
1909         vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1910                 /* found one - only interested if it's shared out of the page
1911                  * cache */
1912                 if (vma->vm_flags & VM_SHARED) {
1913                         i_mmap_unlock_read(inode->i_mapping);
1914                         up_write(&nommu_region_sem);
1915                         return -ETXTBSY; /* not quite true, but near enough */
1916                 }
1917         }
1918 
1919         /* reduce any regions that overlap the dead zone - if in existence,
1920          * these will be pointed to by VMAs that don't overlap the dead zone
1921          *
1922          * we don't check for any regions that start beyond the EOF as there
1923          * shouldn't be any
1924          */
1925         vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1926                 if (!(vma->vm_flags & VM_SHARED))
1927                         continue;
1928 
1929                 region = vma->vm_region;
1930                 r_size = region->vm_top - region->vm_start;
1931                 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1932 
1933                 if (r_top > newsize) {
1934                         region->vm_top -= r_top - newsize;
1935                         if (region->vm_end > region->vm_top)
1936                                 region->vm_end = region->vm_top;
1937                 }
1938         }
1939 
1940         i_mmap_unlock_read(inode->i_mapping);
1941         up_write(&nommu_region_sem);
1942         return 0;
1943 }
1944 
1945 /*
1946  * Initialise sysctl_user_reserve_kbytes.
1947  *
1948  * This is intended to prevent a user from starting a single memory hogging
1949  * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1950  * mode.
1951  *
1952  * The default value is min(3% of free memory, 128MB)
1953  * 128MB is enough to recover with sshd/login, bash, and top/kill.
1954  */
1955 static int __meminit init_user_reserve(void)
1956 {
1957         unsigned long free_kbytes;
1958 
1959         free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1960 
1961         sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1962         return 0;
1963 }
1964 subsys_initcall(init_user_reserve);
1965 
1966 /*
1967  * Initialise sysctl_admin_reserve_kbytes.
1968  *
1969  * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1970  * to log in and kill a memory hogging process.
1971  *
1972  * Systems with more than 256MB will reserve 8MB, enough to recover
1973  * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1974  * only reserve 3% of free pages by default.
1975  */
1976 static int __meminit init_admin_reserve(void)
1977 {
1978         unsigned long free_kbytes;
1979 
1980         free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1981 
1982         sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1983         return 0;
1984 }
1985 subsys_initcall(init_admin_reserve);
1986 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us